Deep Learning Echocardiographic Trajectories of Heart Failure With Preserved Ejection Fraction: A Retrospective...

Heart failure with preserved ejection fraction (HFpEF) is a dynamic chronic disease. Specific disease trajectories remain unclear.

This study aims to investigate cardiac structure and function trajectories measured by echocardiography and their relationship with clinical outcomes.

This was a retrospective cohort from a single center of HFpEF patients with follow-up echocardiogram at ≥ 1 year. Seven longitudinal echocardiographic variables of cardiac geometry and function were utilized to identify trajectory. Longitudinal phenomapping analysis was conducted with machine and deep learning longitudinal clustering. The primary outcome was all-cause mortality, while the secondary outcome consisted of changes in B-type natriuretic peptide levels overtime. A panel of longitudinal laboratory biomarkers was used for exploratory outcomes.

In total, 1,626 HFpEF cases were included; training and validation sets were randomly derived. Echocardiographic trajectories were identified: Echo-Trajectory #1, a stable remodeling trajectory; Echo-Trajectory #2, an increased remodeling course; and Echo-Trajectory #3, decreased remodeling trajectory with right ventricular dysfunction. Baseline clinical characteristics varied significantly among Echo-Trajectories by sex, age, blood pressure, obesity, comorbidities, and left ventricular mass index (P < 0.05). Compared to Echo-Trajectories #2 and #3, Trajectory #1 had a better all-cause mortality outcome (P < 0.001). Similarly, Echo-Trajectory #1 presented favorable trajectories of natriuretic peptides, sodium, and renal function (P < 0.05).

Longitudinal phenomapping resulted in echocardiographic trajectories of cardiac structure and function with different clinical, laboratory, and survival characteristics. HFpEF is a dynamic condition, and longitudinal phenomapping may improve trajectory characterization and guide treatment strategies.